Robust learning algorithm based on agreement among soil sampling techniques

Environmental investigation and modelling require case-based sampling techniques in various domains such as soil, air and water as well as living populations. In most cases, a limited number of sampling techniques can be conducted into a site stemming from the impracticability of geology, time and cost. In addition, if some outliers are recorded due to natural variability and the metrological issues, the modelling process is in need of robust analysis tools. Therefore, a robustness-based sampling agreement and vigorous estimations are needed. The primary purpose of this study is to provide a consensus between different soil sampling methods when a merging is required and to make reliable estimations in case of the existence of any outlier. A machine learning algorithm has been established for reaching targets by considering robustness, transparency, accuracy as well as reproducibility. The algorithm is suited for small data sets and all steps of the algorithm demonstrated that the robust learning algorithm is not severely influenced by the presence of a few outliers. The testing performed based on regression discontinuity analysis and comparative estimations also showed that repeated double robust regression outperforms the conventional multiple least-squares regression. Thus, the learning algorithm can be recommended to the fields of environmental sciences and also may be considered in different disciplines with minor adaptations. (c) 2023 Elsevier B.V. All rights reserved.

Automated summary

Environmental investigation and modelling require case-based sampling techniques in various domains. A machine learning algorithm with robustness, transparency, accuracy, and reproducibility has been established for reaching targets. Testing showed that the algorithm outperforms conventional methods and can be recommended in environmental sciences and other disciplines with minor adaptations.